GAI「詭譎銜接」現象的創作轉化──語意提示詞模組建構與視覺轉譯應用研究

Creative Transformation of GAI“Uncanny Conjunction”Phenomenon: Prompt Keyword Module Construction and Visual Translation Application Research

本研究探討生成式人工智慧（generative artificial intelligence, GAI）影像生成過程中的「詭譎銜接」現象，即影像在異質過渡中的視覺錯位與結構異變特徵。此現象雖源於演算法特性，卻展現獨特的陌生化視覺風格潛力，具備轉化為創作語言的價值。研究旨在將此技術特性轉化為可運用的設計語法資源。透過文獻回顧與大量案例觀察，歸納出「拼貼錯位」、「邏輯錯置」、「異變嵌合」三種GAI異質過渡風格特徵，並據此生成60張GAI影像樣本。經反向語意推導方法和語意群聚分析，建構出四大核心語意提示詞模組分類：「G1拼貼錯位模組」、「G2邏輯錯置模組」、「G3異變嵌合模組」、「G4時空幻擬模組」，確立其為構成GAI詭譎風格的基本語法構件。為驗證此理論架構的應用效能，本研究採用「複合模組」策略進行海報設計之雙重創作實踐驗證：（1）應用於GAI文生圖操作，驗證模組對特定視覺風格的控制效能；（2）作為純文字創作引導語彙，應用於人工創作的設計發想，驗證其作為視覺創意靈感的啟發效能。研究結果顯示，此方法論獲得10位專業人士高度肯定（6.55/7分的滿意度與7.0/7分的使用意願）。證實建構的語意提示詞模組能有效掌控特定視覺風格並作為結構化創作引導語彙，實現從「被動接受」GAI隨機輸出到「主動駕馭與引導」的創作模式轉換，為GAI影像視覺領域提供具體的創作轉化方法論與反向語意推導的新路徑思考。

This study investigates the“uncanny conjunction”phenomenon in generative artificial intelligence (GAI) image generation, characterized by visual dislocations and structural anomalies in heterogeneous transitions. This research aims to transform this technical byproduct into actionable design syntax resources. Through a literature review and extensive case observations, three primary features of GAI heterogeneous transitions were identified:“collage disjunction”,“logical misalignment”, and“morphological fusion”. Based on these identified features, 60 GAI image samples were generated. Using reverse semantic deduction, keywords were extracted from these samples and systematically clustered to construct four core prompt keyword modules (G1-G4), establishing them as fundamental syntactic components of GAI’s uncanny aesthetics. To validate the framework’s practical efficacy, a dual creative practice validation in poster design was conducted using a“composite module”strategy: (1) application in GAI text-to-image operations to verify style control effectiveness, and (2) utilization as textual creative guidance in manual design ideation to verify inspirational value. Findings indicate positive professional evaluations, with a satisfaction score of 6.55/7 and a willingness-to-use score of 7.0/7, confirming that these modules effectively control visual styles and serve as structured creative guidance. This study achieves a paradigm shift from“passive acceptance”to“active mastery”of uncanny aesthetics, providing a concrete creative transformation methodology and an innovative reverse semantic deduction pathway for GAI-based visual creation.